🏏 Cric-360 v2 — Cricket Ground Dataset

13,545 real cricket TV-broadcast ground frames (v1 + v2 combined), plus a 6,418-image synthetic subset available on request for domain-adaptation and augmentation studies.

Built by CogniQube · Lead: Sarim Shahzad · Version 2.0 (July 2026)

v1 and v2 are now combined. This single repo holds every real broadcast frame — the original 3,558 v1 frames and the 9,987 new v2 frames — as one dataset, tagged by a version column.

---

📌 TL;DR — the whole dataset in one table

	grounds (hosted)	games (on request)
What it is	Real cricket TV broadcast frames	Synthetic frames from the Real Cricket mobile game
Images	13,545 (v1: 3,558 + v2: 9,987)	6,418
Availability	✅ Hosted here	📩 On request — see how to get it
Source	Live broadcasts (BBL, IPL, PSL, Tests, bilaterals)	Real Cricket game (Nautilus Mobile / KRAFTON)
Resolution	1280–1924 px wide	2412 × 1080
License	Apache 2.0	Research-use only — ⚠️ not Apache 2.0

Real frames span 35+ venues (v1 + v2 combined), day / night / twilight, many broadcast camera angles.

---

🚀 Quick start

from datasets import load_dataset

# ALL real broadcast frames (v1 + v2 combined = 13,545)
grounds = load_dataset("sarimshahzad/Cric-360", "grounds", split="train")

img  = grounds[0]["image"]     # PIL.Image
meta = grounds[0]              # version, split, quality, source_tag, region, venue ...

▶️ Try it in Google Colab: open the quickstart notebook — no install.

Filter by version or the recommended split (both are metadata columns):

v2_only = grounds.filter(lambda r: r["version"] == "v2")           # 9,987
train   = grounds.filter(lambda r: r["split"]   == "train")        # 10,505
val     = grounds.filter(lambda r: r["split"]   == "validation")   # 1,497
test    = grounds.filter(lambda r: r["split"]   == "test")         # 1,543

Other ways to grab it:

huggingface-cli download sarimshahzad/Cric-360 --repo-type dataset --local-dir Cric-360
git clone https://huggingface.co/datasets/sarimshahzad/Cric-360

---

🧬 v1 + v2 combined

Version	Frames	Location in repo	Notes
v1.0 (Mar 2025)	3,558	data/train|val|test/	Original AI-cleaned release
v2.0 (Jul 2026)	9,987	grounds/	2.8× expansion, richer metadata
Combined	13,545	both	Loaded together as the grounds config

Every row carries a version field (v1 / v2) so you can use the whole set or either release alone.

---

🖼️ grounds — real broadcast frames (13,545)

Genuine frames captured from live cricket TV broadcasts.

Where the v2 frames come from

Source tag	Competition	Region	Images
cognizant	Sponsor-branded series	—	2,424
bbl	Big Bash League	Australia	2,066
frame	Unspecified broadcast	—	1,449
test	Test match	—	1,090
irl vs ind	Bilateral series	Ireland v India	826
psl / psll	Pakistan Super League	Pakistan / UAE	755
ipl	Indian Premier League	India	553
ban test	Test match	Bangladesh	408
sunrise	Franchise T20	—	330
ind vs afg, ban, aus	International	—	86

v1 frames carry source_tag = v1_broadcast and, where available, a venue and has_carpet flag. Source tags are parsed from filenames; unconfirmed competitions/regions are marked "Unspecified".

Resolution & quality (v2 frames)

Quality tier	Definition	Images
HD	≥ 1920 px wide (1080p-class)	3,555
HD-Ready	1280–1919 px wide (720p-class)	6,432

All JPEG, RGB. Width 1280–1924 px, height 602–990 px.

---

🎮 games subset (synthetic, 6,418 images) — available on request ⚠️

📩 This subset is not hosted here. It is shared on demand for research use. To request it, open an issue on the GitHub repo or contact CogniQube. A metadata/games_metadata.csv describing all 6,418 frames is available in the GitHub repository so you can inspect its contents before requesting.

Screenshots captured from the Real Cricket mobile game by Nautilus Mobile (published by KRAFTON). These are computer-generated (CGI) frames — rendered players, crowds, stadiums and in-game UI — not real matches.

Property	Value
Images	6,418
Resolution	2412 × 1080 (fixed, Full-HD mobile)
Rendering	Real-time 3D game engine (synthetic)
Visible branding	Real Cricket, Nautilus Mobile, KRAFTON, WCA boundary boards

Why synthetic frames? A controlled, high-resolution complement to noisy real footage — useful for data augmentation, pre-training, and sim-to-real / domain-adaptation experiments.

⚠️ These images depict a copyrighted third-party product. They are provided for non-commercial research use only and are excluded from the Apache 2.0 license. All rights to the Real Cricket game remain with Nautilus Mobile / KRAFTON. Rights-holders may open an issue on GitHub.

---

🌍 Diversity

Dimension	Coverage
Venues	35+ stadiums worldwide across v1 + v2 (Australia, India, Pakistan, UAE, Bangladesh, Ireland, and more)
Lighting	Daylight · day–night twilight · full floodlit night · overcast
Camera angles	Wide broadcast, side-on, straight, high/overhead, low-angle, on-ground
Ground state	Fresh & worn pitch, footmarks, shadows (player / stand / floodlight), partial occlusion
Formats	T20 leagues, ODIs, Test matches, bilateral internationals

🎯 What you can build with it

Task	How the data helps
Ground / pitch segmentation	Wide variety of turf, wear, and lighting — v1 masks available (below)
Player & ball detection / tracking	Many players-on-field frames across angles
Homography & camera calibration	Consistent broadcast geometry, known pitch dimensions
AR / virtual ad insertion	Boundary boards, sightscreens, empty-ground regions
Depth & scene understanding	Wide depth range from foreground pitch to far stands
Sim-to-real / domain adaptation	Pair the real grounds set with the synthetic games set

🧩 v1 Ground-Segmentation masks are available

Pixel-level ground-segmentation masks for the v1 frames are available (they label the playable turf region) — ideal for training and benchmarking segmentation models. Request them via the GitHub repo. Masks for the v2 frames are on the roadmap.

---

🗂️ Files & metadata schema

Cric-360/                  (Hugging Face)
├── README.md              ← this card
├── metadata.csv           ← combined export of all 13,545 rows (human reference)
├── dataset_stats.json
├── notebooks/Cric360_v2_quickstart.ipynb
├── grounds/               ← v2 real frames (9,987 .jpg) + metadata.csv
└── data/                  ← v1 real frames (3,558 .jpg in train/ val/ test/) + metadata.csv

# The loader reads the per-folder grounds/metadata.csv and data/metadata.csv (same schema),
# so `load_dataset(..., "grounds")` returns all 13,545 frames with metadata attached.
# games/ (6,418 synthetic frames) is shared on request — its metadata.csv is in the GitHub repo.

Every image row (same schema in both folders) has:

Column	Description
file_name	Path relative to repo root (grounds/… or data/…)
version	v1 or v2
subset	grounds (real)
content_type	broadcast
split	Recommended train / validation / test
width, height, resolution, aspect_ratio	Image geometry
quality	HD, HD-Ready, Full HD
image_mode, format	RGB, JPEG
source_tag, competition, region	Origin labels (v2); v1 uses v1_broadcast
venue, has_carpet	v1 labels where available

---

📜 Licensing

Part	License / terms
grounds real frames (v1 + v2), metadata, code	Apache 2.0 — research & commercial use with attribution
games synthetic frames	Research use only — © Nautilus Mobile / KRAFTON, not Apache 2.0

Underlying broadcast footage remains the property of the respective rights-holders; real frames are shared under Apache 2.0 for research, consistent with the v1 release.

📖 Citation

@dataset{shahzad_cric360_v2_2026,
  author  = {Shahzad, Sarim and {CogniQube}},
  title   = {Cric-360 v2: A Cricket Broadcast and Synthetic Ground Dataset for Computer Vision},
  year    = {2026},
  version = {2.0},
  url     = {https://huggingface.co/datasets/sarimshahzad/Cric-360}
}

🙏 Acknowledgements

• CogniQube — dataset construction & curation (github.com/Cogni-Qube)
• Broadcast rights-holders of the respective competitions (BBL, IPL, PSL, and others)
• Nautilus Mobile / KRAFTON — creators of the Real Cricket game shown in the synthetic subset

Questions, corrections, or rights concerns → open an issue on GitHub.